Switch from the very slow os.path.splitext to str.rpartition.

As a consequence, `ext` doesn't include the leading dot anymore, so
change `suffixes` accordingly too.

importlib_metadata/__init__.py

@@ -493,15 +493,17 @@ class Prepared:
     """
 
     normalized = None
-    suffixes = '.dist-info', '.egg-info'
+    suffixes = 'dist-info', 'egg-info'
     exact_matches = [''][:0]
 
     def __init__(self, name):
         self.name = name
         if name is None:
             return
         self.normalized = self.normalize(name)
-        self.exact_matches = [self.normalized + suffix for suffix in self.suffixes]
+        self.exact_matches = [
+            self.normalized + '.' + suffix for suffix in self.suffixes
+        ]
 
     @staticmethod
     def normalize(name):
@@ -520,8 +522,10 @@ def legacy_normalize(name):
 
     def matches(self, cand, base):
         low = cand.lower()
-        pre, ext = os.path.splitext(low)
-        name, sep, rest = pre.partition('-')
+        # rpartition is like os.path.splitext, but much faster.  They'd only
+        # differ if pre is empty, but in that case we don't have a match anyways.
+        pre, _, ext = low.rpartition('.')
+        name, _, rest = pre.partition('-')
         return (
             low in self.exact_matches
             or ext in self.suffixes